Patriot EP 128GB: fast for 4 GB, then under its own V30 floor

What is the Patriot EP?

Patriot Memory is a memory brand from Fremont, California, best known for PC memory and its Viper gaming line. This is its EP series microSD card in the V30 A1 version, part number PEF128GEP31MCX, and the box says Made in Taiwan beside a Designed in California seal. Patriot also sells cards called EP Pro and EP Express; they are different products, I tested only this one, and nothing below carries over to them. I bought it sealed on Amazon, and it is the first Patriot card on my bench.

It is worth testing because of what the box leaves out. It prints no read or write speed at all. What it does print, on the back and in eight languages, is 30 Megabytes per second minimum sustained write speed, which is the V30 mark spelled out in words. Patriot’s own product page and its spec sheet for this part number add the speeds the box omits: up to 90 MB/s read and 80 MB/s write. The card is in CKBench now, the running index of every card on my bench, next to the other budget 128 GB cards I have bought this autumn, among them the Silicon Power Superior Pro, which turns out to share its registers, and the LinkMore XV23 Lite.

What the Patriot EP claims, and what it did

Because the box prints no speeds, my rubric takes the ones Patriot publishes for this exact part number, and its spec sheet carries the same UPC as the box. The read lands: 97.2 MB/s is where every standard UHS-I card on my bench lands, the practical ceiling of the SDR104 mode, and it clears Patriot’s 90. The write lands too, but only on a new card and only while its fast stage lasts; the next sections show how short that is. The line that matters is the one the box does print. A 30 MB/s minimum sustained write is exactly what the V30 and U3 marks mean, and the card sustained 22.54 MB/s, a quarter short, on both readers and in every state I put it in. A1 cannot be settled on a USB reader; the card cleared its floors at queue depth one, which is indicative and nothing more.

A blister with no speeds and one promise

The EP comes sealed in a retail blister with a full-size SD adapter. The front carries every mark, V30, A1, UHS-I, U3 and Class 10, plus “Supports 4K”, and no speed. The back has the line this review turns on, 30 Megabytes per second minimum sustained write speed, in English, French, Spanish, Portuguese, Italian, Russian, German, Japanese and two forms of Chinese; the Portuguese version says megabits, a translation slip. It also carries a limited three-year warranty badge, Windows 10 and Mac badges, the footnote that capacities are unformatted and that 1 GB equals a billion bytes, and Patriot’s sticker with the part number and UPC. The Taiwanese compliance label calls it a microSDHC card, boilerplate from an older package; it is microSDXC. There was no returns sticker on any panel.

The card arrived empty and formatted exFAT with 128 KB clusters and no volume label, a clean factory layout. Its volume serial number, which an ordinary format fills with a value derived from the clock, read 0E16-2000. That is no accident: written as one number, 0E162000 is the length of the card’s partition in sectors, so whatever formatted it at the factory used the partition size as the serial. Hold that thought, because one other card on my bench arrived with a serial built the same way.

Five minutes at 22.7 MB/s, under the V30 floor

The timed write starts after an untimed warm-up that runs until the speed stops changing, so the five minutes that decide the video class start on a settled card. On this card the warm-up began at 8 to 12 MB/s for its first eight seconds, the card still cleaning up after the random-write test, then sat at 22 to 23 and settled at 90 seconds. The trace below is the scored run, one sample a second, with the speed-class floors marked. The first thirty seconds averaged 22.65 MB/s and the last thirty 22.75, the whole five minutes 22.74, and the worst ten seconds were 22.54 MB/s. The longest single write took 112 milliseconds; the bench flags anything over half a second. On the second reader, twice, the same five minutes ran at 22.68 steady with worst ten seconds of 22.44.

22.54 is 7.46 MB/s under the V30 floor, 75% of it, and U3 carries the same 30 MB/s floor, so both marks fail; V10 holds with more than twice its floor in hand. There is nothing dramatic in the trace to blame: no dips, no stall, no cliff, just a card whose own write speed is 22.7 MB/s. In the CKBench table, only three no-brand fakes hold a lower floor. For context, the SUNEAST Ultimate Pro lost V30 only after it had been filled, at 27.4; this card never had it.

At least 4.4 GB at 91 MB/s when new, about 1 GB after its first full card

The fast writes are real, for a while. Like many budget cards, this one writes new data into a fast area first and moves it into slower flash afterwards. On a new card, every write before the random-write test ran at 88 to 91 MB/s: a 64 MB check, a 1 GiB preparation file and three 1 GiB passes, 4.36 GB in all. The fast stage never ran out inside that, so how far beyond 4.36 GB it would have gone, I cannot tell you; the suite pauses there. By the time the timed write began, the random test and the warm-up had used it up.

Then I filled the card once, end to end, and emptied it. On the next run, the 1 GiB preparation file took 17.1 seconds, which means about 1 GB of it went at 90 MB/s and the rest at 22.8, and all three 1 GiB passes after it ran at 22.8. The card had sat idle for 19 hours first, so it was not still busy. A card reader cannot tell a card which space is free again, and that is the usual explanation for a card that slows down after a fill; I wrote it up in why a used microSD card tests slower. So I sent the strongest reset my bench can: from a Raspberry Pi’s native SD slot, a discard of all 121 GB, the command that tells a card every block is free, then a fresh format. The next run took about 1.1 GB fast and wrote 22.8 again. I cannot tell from outside whether the card ignored the discard or whether freeing the space simply does not bring its fast stage back; either way, for a buyer, the 4 GB burst belongs to a new card only.

That is also the honest reading of Patriot’s 80 MB/s write figure. On a new card it held at 114%; after one full card, the same test ran at 22.8, 28.5% of it. Its register twin, the Silicon Power Superior Pro, keeps its fast stage after a fill: at least 4.36 GB both times, and 4.8 and 4.7 GB more after its random test. The SUNEAST Ultimate Pro loses most of its much larger one.

121.0 GB, every byte verified, and only 94.55% of the label

The capacity test wrote 113 files of unpredictable, dedup-proof data until the card was full, 120,984,043,520 bytes, then read every one of them back and compared. Zero mismatched bytes. Zero read errors. At this card’s write speed the fill took 89 minutes 29 seconds; the read-back took 21 minutes 27 seconds at 98.3 MB/s. Across the first, middle and last tenth of the card the write ran at 24.4, 22.6 and 22.6 MB/s, the first tenth lifted by a short fast stage at the start, and the read at 94.2, 94.2 and 94.1.

The fill is the clearest picture of this card. After a pause before it began, the card wrote about 1.2 GB at 87 MB/s, then dropped to 22.6 and stayed there for the rest of the card: 99.7% of the fill’s seconds ran under 30 MB/s, and its worst ten seconds were 20.76. The Silicon Power Superior Pro, filled on the same reader, spent one second of its 3,130 under 30, and its worst ten seconds there were 36.6.

Every byte it declares is real. What it declares is the smaller story: 121,018,253,312 bytes, 94.55% of the 128 GB on its label, about 7 GB short of it. Every other genuine card I have tested declares between 96.7% and 100.1% of its label; the box’s footnotes, that capacities are unformatted and a gigabyte is a billion bytes, do not cover a gap this size. Windows will show the card as about 112.7 GB, because it counts in binary; that part is arithmetic, and the methodology page has the conversion. The capacity row of the rubric scores the declared share of the label, and 94.55% is a 3.

The Silicon Power Superior Pro’s registers, under a blank name

Every SD card carries a set of registers, written at the factory, that a USB reader cannot show you and a native SD host can. One names the card: manufacturer ID, OEM ID, product name, revision, serial and build date. The others describe it: how big it is, which SD specification it follows, which speed classes it declares. I read this card’s on a Raspberry Pi before its first run. Its naming fields are nearly empty, and its describing fields match a card I reviewed in September, the Silicon Power Superior Pro 128GB, almost byte for byte.

The manufacturer and OEM IDs are blank, so the card does not say who built it; Patriot’s own ID in the public lists is not used. The product name, Viper, is Patriot’s gaming brand. Everything else lines up with the Superior Pro, down to a reserved field no other card on my bench sets to 7 and to the factory serials, which on both cards are the partition’s length in sectors (0x0E7B6000 is the Superior Pro’s). That points to the same controller family set up for two brands, the app class matching each brand’s print. It does not tell you whose flash is inside, and my manufacturer ID register has more on what these fields can and cannot prove. The laser marks on the card back read like a work order with a date in it, 22 November 2025, seven months before the register’s build date, so I would not read anything into them.

Then I ran them on the same reader. The Superior Pro writes about 39 MB/s after its fast stage and holds V30 with a quarter in hand; the Patriot writes 22.7, 58% of that, and holds V10. The Superior Pro keeps its fast stage after a fill; the Patriot loses most of it. Same register platform, a slower card: it is the second time this autumn that matching registers have hidden a different card, after the LinkMore XV23 Lite and the SUNEAST.

One random-write pass fell to 0.72 MB/s

Loop recording is a long write interrupted constantly by small ones, and random 4K writes are the bench version of that: three twenty-second passes, and the question is whether any later pass falls below the first. On this card one did, hard. The scored run did 1,125, 176 and 1,151 write IOPS, 4.61, 0.72 and 4.71 MB/s: the middle pass fell to 16% of the first, and the third recovered.

It was not a one-off. On the second reader the third pass fell instead, to 2.61 MB/s, 54% of the first, and after the discard and format the third pass fell to 3.71, 76% of the first. Afterwards the card was back to 90% of its sustained rate 17 seconds into continuous writing, after starting the warm-up at 8 to 12 MB/s.

This card changed how the rubric reads the test. Until protocol v1.28 the endurance row compared the first pass with the third, which on this run reads as no fade at all, since the third pass was the fastest of the three; the deepest drop on the scored run fell in the one pass the old rule did not look at. Since 21 October 2026 the row compares the first pass with the slowest pass after it, and on this card that is 84.38%, the bottom row of the table. A pass that falls to a sixth of the first is exactly the stall a loop recorder runs into, and a rule that cannot see it is the wrong rule.

135 °F at most, and no throttling

Patriot rates the card for -25 to 85 °C, on its product page and spec sheet rather than the box. At the end of five minutes of uncooled writing in a 76 °F room the card read 134 °F on the scored run and 135 °F on both runs on the second reader. The rubric takes the highest reading on file: 135 °F is 57.2 °C, 27.8 °C under Patriot’s maximum, which scores a 5.

It did not throttle: the last thirty seconds of the timed write ran at 22.75 MB/s against 22.65 for the first thirty. Heat follows work, and this card does little of it; the LinkMore XV23 Lite, writing at more than three times the rate on the same reader, reached 153.5 °F. With 27.8 °C of margin, heat is the one thing on this card that is not a concern.

Three runs, two readers, one floor

Every card I review runs on my primary reader, a ProGrade PGM0.5, and again on a SanDisk QuickFlow SDDR-B751. This card ran once on the ProGrade, with its whole-card fill, and twice on the QuickFlow: once after the fill, and once after the discard and the format.

The floor does not move: 22.54 and 22.44 MB/s, new and used, on either reader, within half a percent. What moves is the fast stage, and that follows the card’s history, not the reader, since the Superior Pro kept its fast stage on the same QuickFlow. Random reads always run higher on the QuickFlow, which is normal on this bench.

How the 2.5 breaks down

Under the rubric as it stood before 21 October, the endurance row would have read 5 and the overall 3.5. I changed the rule because of this card, and the change moved two other published cards’ endurance rows by half a point with no change to their overall scores.

The bottom line

The Patriot EP 128GB holds every byte it declares, and almost everything else on its box overstates it. The V30 and U3 marks, and the 30 MB/s minimum sustained write the box spells out, are missed by a quarter on two readers and in every state I could put it in. Patriot’s 80 MB/s write figure holds only for a new card’s first few gigabytes, and the space it declares is 94.55% of the label. For a dash cam at ordinary bitrates, or for photos and music on a phone, it works. For anything that needs the marks it prints, it does not, and the Silicon Power Superior Pro, on the same register platform, holds V30.

It is also the second card this autumn whose fast stage did not survive its first full card, and the first where I could send the card a discard from a native host to try to bring it back. It did not come back. If a card’s headline speed is a cache, the size of that cache after a fill is worth knowing before you buy it, and the box will not tell you.

Sample size: one unit, purchased sealed at retail from Amazon. Three runs on two readers on 16, 17 and 21 October 2026, scored under CKBench protocol v1.28.

Disclosure: I bought this card at retail from Amazon. Patriot Memory had no involvement in or knowledge of this review, saw no result before publication, and had no opportunity to influence any number on this page. This post contains affiliate links. As an Amazon Associate I earn from qualifying purchases, at no extra cost to you. Affiliate links like these and my YouTube channel are what keep the site running.